Repository navigation
Description
Summary
The e2e playbooks are strong at proving data equality and weak at proving what the controllers report and what they clean up. Regressions in status/conditions or temporary-resource leaks pass every playbook.
Current assertion coverage across the 38 test-e2e/test_*.yml playbooks:
| Assertion | Playbooks |
|---|---|
status.latestMoverStatus |
22/38 |
status.lastSyncTime |
11/38 |
status.lastSyncDuration |
10/38 |
status.conditions |
1/38 (test_restic_without_trigger.yml) |
resource cleanup (state: absent checks) |
3/38 |
So Synchronizing=False with the expected reason, condition message sanity, and "temporary PVC/Snapshot/Job/Service is gone after cleanup" are effectively untested end to end. These are exactly the signals users and downstream operators (ACM/ODF) monitor, and the leaks that show up as support cases.
Proposal
Add one reusable role (e.g. roles/verify_sync_status) and call it from playbooks at points where they already wait for data, so the assertions cost nothing beyond the in-cluster kubectl get they replace:
- Source/destination status:
status.lastSyncTimeadvanced past the sync start;status.lastSyncDurationset and plausible;status.nextSyncTimeset when a schedule is configured. - Conditions:
SynchronizingisFalsewith the expected reason, and no staleTruecondition is left behind after completion. - Mover status:
latestMoverStatus.result == "Successful"(and on a negative-path playbook, the error text/condition reason). - Cleanup: after completion, no mover Job/Pod remains, and temporary PVCs/
VolumeSnapshots created for the copy method are gone — except the protected/latest image, which must still exist. - Idempotence spot check: a second reconcile/no-op cycle does not create duplicate or renamed resources.
Keep the existing data-comparison roles as-is; this is additive and runs against the same cluster state.
Acceptance criteria
- A single shared role implements the status/condition/cleanup assertions; playbooks do not grow bespoke
until:blocks. - All playbooks assert at least the completion condition (
Synchronizing=False) and cleanup of temporary resources. - At least one playbook (can be a new negative-path case, see e2e: cover paused, Clone copy method, CLI migration, and a negative path #2190) asserts
Synchronizing=Falsewith an error reason. - No measurable wall-clock increase: assertions run in-process against objects the playbooks already wait on; no new cluster, job, or image.
(AI generated)
Metadata
Metadata
Assignees
Labels
Type
Projects
- StatusShow more project fieldsDone
Activity